Invalid Results-way too many

Forums › Cruncher's Corner

Anonymous
9 Sep 2005 0:45:14 UTC
Topic 13116
(moderation:
    )

    Quote:

    Anyway, a response from BA or BM would be much appreciated.

    Gary, you spend so much time solving other peoples problems, that I thought I had a duty to reply!

    Currently the validator marks about 0.3% of results INVALID. Most of these are probably valid, it's just that the validator has a difficult job to do and is sometimes too conservative. We've done our best to tune it, but at the 0.3% level we don't know how to improve it.

    The different numerical results returned by different CPUs, different OSes and different OS versions all mean that sometimes results are correct, but not recognized as such.

    I looked to see what was going on with jrenkar's results, but all the ones currently in the database are VALID. No sign of trouble.

    Cheers,
    Bruce

    Bruce Allen
    Bruce Allen
    Joined: 15 Oct 04
    Posts: 958
    Credit: 170,849,008
    RAC: 0

    Invalid Results-way too many

    11 Sep 2005 8:58:39 UTC
    Message 2700
    (moderation:
      )

      Quote:
      Bruce me thinks if you do a newer pole you will find that your 0.3% is rising


      It's not rising. Look at the server status page, linked from Einstein@Home front page.

      Quote:
      I also urge you into looking into the recent examples of work unit ID's given on this post as you can still access,and as I believe the more examples we give you the better you are to solve the problem :)


      Currently none of your machines have any invalid results for me to look at.

      It is indeed the case that results from different OSes are more likely to disagree. So this leads to higher invalid rates for Linux and Mac OSX users, because there are fewer of them than Win32 users. But even for Mac OSX users (the minority at E@H) there are currently 16726 valid results in the database and 221 invalid results. This is an invalid rate of only about 1.3%, which is low enough for me not to be too concerned, even if most of these are due to the validator being too conservative.

      Thanks for contributing to E@H. Please don't give up on us!

      Cheers,
      Bruce

      Bruce Allen
      Bruce Allen
      Joined: 15 Oct 04
      Posts: 958
      Credit: 170,849,008
      RAC: 0

      RE: RE: But even for Mac

      12 Sep 2005 5:50:28 UTC
      Message 2701
      (moderation:
        )

        Quote:
        Quote:
        But even for Mac OSX users (the minority at E@H) there are currently 16726 valid results in the database and 221 invalid results. This is an invalid rate of only about 1.3%, which is low enough for me not to be too concerned, even if most of these are due to the validator being too conservative.

        Bruce, 1.3% wouldn't bother me a bit... but I'm concerned that the figure might be much higher for certain CPUs within the "Mac OSX" group. All of the recent optimizations were for Altivec - I would expect if there are any problems, it would show up most on 7450/7455 G4s, second most on G5s and 7400/7410 G4s, not at all on other CPUs.

        Obviously my G4 (7455) can do far more Einstein WUs than my G3, and I'll certainly leave it crunching - but I'll also keep an eye on the results, and if the "invalid" percentage gets TOO bad, then I'll just "shuffle" the Einstein work around to another box and suspend it on that one.

        We won't know for some time - until at least 50 or so WUs have been returned from my G4, I'd consider any numbers as "statistically questionable".

        Bill, also keep in mind that the 1.3% is an average across ALL workunits. The results do depend upon the data set; it may be that there are some data sets that tend to produce larger differences across processors and/or architectures. For these specific workunits, the rate may be higher.

        Bruce

        Bruce Allen
        Bruce Allen
        Joined: 15 Oct 04
        Posts: 958
        Credit: 170,849,008
        RAC: 0

        RE: I'm sorry for

        20 Sep 2005 14:18:07 UTC
        Message 2702
        (moderation:
          )

          Quote:
          I'm sorry for resurrecting this thread which had gone dormant but I felt I needed to add my own postscript to it. There are a number of points I wanted to make.
          [...SNIP...]
          My personal prediction is that as more people visit the website and discover this report, there will be a steady stream of previously "diehard" Seti only crunchers who will suddenly realise that EAH is equally worthy of their support. As one of those who were canvassing for this type of progress report many months ago, let me say that the wait has been more than worthwhile. Well done - from a very satisfied contributor.

          Gary, thank you!

          Comment viewing options

          Select your preferred way to display the comments and click "Save settings" to activate your changes.

          Forums › Cruncher's Corner